基于神经网络的重构指令预取机制及其可扩展架构

doi:10.3969/j.issn.0372-2112.2012.07.031

电子学报

基于神经网络的重构指令预取机制及其可扩展架构

陈志坚, 孟建熠, 严晓浪, 沙子岩

浙江大学超大规模集成电路设计研究所, 浙江杭州 310027

收稿日期:2010-03-30 修回日期:2012-03-01 出版日期:2012-07-25
- 作者简介:
- 陈志坚男,1984年7月出生于浙江洞头,2006年获浙江大学自动化学士学位,现为浙江大学超大规模集成电路设计研究所硕博连读生,主要研究方向为高性能嵌入式处理器设计,可重构处理器设计. E-mail:chenzj@vlsi.zju.edu.cn 孟建熠男,讲师,1982年1月出生于浙江上虞,分别于2004年和2009获浙江大学电子信息工程学士和电路与系统博士,主要研究方向为高性能低功耗嵌入式处理器设计. E-mail:mengjy@vlsi.zju.edu.cn

Algorithm/Architecture of NN-Based Configuration Prefetching

CHEN Zhi-jian, MENG Jian-yi, YAN Xiao-lang, SHA Zi-yan

Institute of VLSI Design Zhejiang University, Hangzhou, Zhejiang 310027, China

Received:2010-03-30 Revised:2012-03-01 Online:2012-07-25 Published:2012-07-25

摘要/Abstract

摘要： 针对动态可重构处理器的配置信息加载延时,提出了一种基于神经网络的可扩展的重构指令预取机制.增加感受器的历史指令信息,并结合感受器权重构建新型的感受器模型,通过权重与历史指令信息的协同训练学习重构指令调用规律.在处理器运行过程中,提前完成对后续重构指令的预测及配置信息的预取,隐藏指令重构成本.进一步提出了本方法的可扩展实现框架,神经网络的学习结果作为重构指令的关联信息,被移至内存并分布式存储.在重构指令预取时,完成对神经网络学习信息的加载.实验结果表明,该方法对重构指令的预测准确率达91%,综合性能平均提升40%.

关键词: 可重构处理器, 配置信息预取, 改进神经网络算法, 可扩展存储架构

Abstract: Reconfigurable processor suffers a severely performance loss from reconfiguration overhead.A NN(neural network algorithm) based configuration prefetching algorithm was proposed in this paper to reduce the overhead.Not only the receptor weight but also the history RFUOP ID constructs an advanced receptor model.The neural network studies the RFUOP trace through collaborative training of receptor weight and history RFUOP ID.With the learning result,the neural network predicts next RFUOPs and completes configuration prefetching,overlapping the configuration loading with the computation on the host processor.Furthermore,an extensible architecture for neural network storage was proposed.As a component of RFUOP,reception information is located on off-chip memory and linked to corresponding configuration field.Once configuration was prefetched,neuron information was loaded and computed for next prediction.Experiments show that the NN-based prefetching algorithm can reach 91% prediction accuracy,while gaining performance improving by 40% on average.

Key words: reconfigurable processor, configuration prefetch, advanced neural network algorithm, extensible architecture

中图分类号:

TN302

陈志坚, 孟建熠, 严晓浪, 等. 基于神经网络的重构指令预取机制及其可扩展架构[J]. 电子学报, DOI: 10.3969/j.issn.0372-2112.2012.07.031.

CHEN Zhi-jian, MENG Jian-yi, YAN Xiao-lang, et al. Algorithm/Architecture of NN-Based Configuration Prefetching[J]. Acta Electronica Sinica, DOI: 10.3969/j.issn.0372-2112.2012.07.031.

参考文献

[1] A Lodi,M Toma,et al.A VLIW processor with reconfigurable instruction set for embedded applications [J].IEEE Journal of Solid-State Circuits,2003,38(11):1876-1886.

[2] 孟建熠,丁永林,葛海通,等.一种RTL级数据通路ODC低功耗优化算法[J].电子学报,2010,38(7):1654-1659. Meng Jian-yi,Ding Yong-lin,Ge Hai-tong,et al.A RTL level ODC algorithm for data path low power optimization [J].Acta Electronica Sinica,2010,38(7):1654-1659.(in Chinese)

[3] 宋宇鲲,高明伦,邓红辉,等.DReAC:一种新型动态可重构协处理器[J].电子学报,2007,35(5):833-837. Song Yun-kun,Gao Ming-lun,Deng Hong-hui,et al.DReAC:A novel dynamically reconfigurable Co-processor [J].Acta Electronica Sinica,2007,35(5):833-837.(in Chinese)

[4] M J Wirthlin,B L Hutchings.Dynamic instruction set computer.Proceedings of the IEEE Symposium on FPGA's for Custom Computing Machines.Napa Valley,CA,USA:IEEE Computer Society,1995.99-107.

[5] A Dandalis,V K Prasanna.Configuration compression for FPGA-based embedded systems [J].IEEE Transactions on Very Large Scale Integration (VLSI) Systems,2005,13(12):1394-1398.

[6] U Malik,O Diessel.On the placement and granularity of FPGA configuration.Proceedings of the IEEE International Conference on Field-Programmable Technology.Brisbane,Australia:IEEE Electron Devices Society,2004.161-168.

[7] Z Li,K Compton,S Hauck.Configuration cache management techniques for reconfigurable computing.Proceedings of the IEEE Symposium on Field-Programmable Custom Computing Machines.Washington,DC,USA:IEEE Computer Society,2000.22-36.

[8] Q Yang,J-P Soininen,et al.A configuration locking technique to reduce the configuration overhead of run-time reconfigurable devices.Proceedings of the International Symposium on System-on-Chip.Tampere,Finland,2007.1-5.

[9] J Resano,D Mozos,et al.A hybrid prefetch scheduling heuristic to minimize at run-time the reconfiguration overhead of dynamically reconfigurable hardware.Proceedings of the Conference on Design,Automation and Test in Europe.Washington,DC,USA:IEEE Computer Society,2006.106-111.

[10] J RResano,et al.Efficiently scheduling runtime reconfigurations [J].ACM Transactions on Design Automation of Electronic Systems,2008,13(4):58-65.

[11] B B Wu,et al.Run-time configuration prefetching to reduce the overhead of dynamically reconfiguration.IEEE International SoC Conference.Las Vegas,NV,USA:IEEE Press,2010.305-308.

[12] X Tang,M Aalsma,R Jou.Compiler directed approach to hiding configuration latency in Chameleon processors.Proceedings of the Roadmap to Reconfigurable Computing,10th International Workshop on Field-Programmable Logic and Applications.London,UK:Springer-Verlag,2000.29-38.

[13] Z Li,S Hauch.Configuration prefetching techniques for partial reconfigurable coprocessor with relocation and defragmentation.ACM/SIGDA International Symposium on Field Programmable Gate Arrays.Monterey,CA,United states:Association for Computing Machinery,2002.187-195.

[14] D Jimenez,C Lin.Neural methods for dynamic branch prediction [J].ACM Transactions on Computer Systems,2002,20(4):369-397.

[15] C-SKY High Performance Embedded Processor.http://www.c-sky.com,2008.

基于神经网络的重构指令预取机制及其可扩展架构

Algorithm/Architecture of NN-Based Configuration Prefetching

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 0

编辑推荐

Metrics